67f0c3491a4a1966603ace4fe110b7cfc5b64e9c | Author: Alexander Dejanovski <alex@thelastpickle.com>
| 2020-11-17 10:18:22-08:00
SSTableLoader will fail if encryption parameters are used due to CASSANDRA-16144
patch by Alexander Dejanovski; reviewed by David Capwell, Jon Meredith for CASSANDRA-16280
6fc8920889e8537a1f56f45e6c966b3d18325fbb | Author: Alexander Dejanovski <alex@thelastpickle.com>
| 2020-06-26 11:08:50+02:00
Fix upgrade failures when using the EC2Snitch in legacy mode
Legacy naming conventions for AWS regions were allowing datacenter names such as us-west and us-west-2, which in the new standard mode become us-west-1 and us-west-2.
As us-west-2 can match both the standard and legacy mode, checking the DC name cannot be reliably used to detect mixed modes being used as it would fail in case of an upgrade from pre-4.0 clusters in a region such as us-west-2.
The rack check should be enough as rack names in the standard mode now includes the region name, which can be easily identified.
patch by Alexander Dejanovski; reviewed by Mick Semb Wever, Joey Lynch for CASSANDRA-15878
dd9a6e0a8a691bdee357eb6559f1b2b92b076925 | Author: Jon Haddad jon@jonhaddad.com <jon@jonhaddad.com>
| 2020-03-24 13:44:17-07:00
Use more appropriate logging levels.
The following logging was changed:
* When the Commit log reader finishes logging each segment, from debug to info
* When starting a new compaction, logging details of the compaction task, from debug -> info
* When finishing a compaction, logging details of how many sstables, sizes, etc, from debug -> info
* When performing a validation compaction, adding helpful logging
* When flushing memtables, log details. debug -> info
* When flushing largest memtable to free up room, logging more details at INFO
* Various logging messages for incremental repair moved to INFO from debug
* When failing incremental repair phase, log a warning instead of debug message
Patch by Jon Haddad; Reviewed by Alexander Dejanovski for CASSANDRA-15661
9d649d69a56a91fcb06a3582b22606f0fe361f49 | Author: Jon Haddad <jon@jonhaddad.com>
| 2018-02-08 11:01:38-08:00
Count deleted rows scanned during reads for tracing and tombstone thresholds.
If a row is read but is not live anymore (which happens with row level tombstones) it is not counted anywhere
in the metrics nor reported in tracing. Row tombstones themselves are not reported anywhere.
The consequence is that some delete heavy workloads will show no tombstone read but endure severe
performance issues. This commit counts deleted rows as standard tombstone cells.
Patch by Alexander Dejanovski; Reviewed by Jon Haddad for CASSANDRA-8527
419552a93f4e164dfce3cb6fea15328abade5d3f | Author: Alexander Dejanovski <alex@thelastpickle.com>
| 2017-12-18 18:27:44+01:00
Count deleted rows scanned during reads for tracing and warning tombstone thresholds.
If a row is read but is not live anymore (which happens with row level tombstones) it is not counted anywhere
in the metrics nor reported in tracing. Row tombstones themselves are not reported anywhere.
The consequence is that some delete heavy workloads will show no tombstone read but endure severe
performance issues. This commit counts deleted rows as standard tombstone cells.